python - tensorflow map_fn TensorArray has inconsistent shapes -

i playing around map_fn function, , noticed outputs tensorarray, should mean capable of outputting "jagged" tensors (where tensors on inside have different first dimensions.

i tried see in action code:

import tensorflow tf import numpy np  num_arrays = 1000 max_length = 1000  lengths = tf.placeholder(tf.int32) tarray = tf.map_fn(lambda x: tf.random_normal((x,),0,1),lengths,dtype=tf.float32) #should return tensorarray  # starttensor =  tf.random_normal(((tf.reduce_sum(lengths)),),0,1) # tarray = tf.tensorarray(tf.float32,num_arrays) # tarray = tarray.split(starttensor,lengths) # outarray = tarray.concat()   tf.session() sess:     outputarray,l = sess.run([tarray,lengths],feed_dict={lengths:np.random.randint(max_length,size=(num_arrays))})     print outputarray.shape, l 

however got error:

"tensorarray has inconsistent shapes. index 0 has shape: [259] index 1 has shape: [773]"

this of course comes surprise me since under impression tensorarrays should able handle it. wrong?

while tf.map_fn() use tf.tensorarray objects internally, , tf.tensorarray can hold objects of different size, program won't work as-is because tf.map_fn() converts tf.tensorarray result tf.tensor stacking elements together, , operation fails.

you can implement tf.tensorarray-based using lower-lever tf.while_loop() op instead:

lengths = tf.placeholder(tf.int32) num_elems = tf.shape(lengths)[0] init_array = tf.tensorarray(tf.float32, size=num_elems)  def loop_body(i, ta):   return + 1, ta.write(i, tf.random_normal((lengths[i],), 0, 1))  _, result_array = tf.while_loop(     lambda i, ta: < num_elems, loop_body, [0, init_array])